Agents as the only developers
General Intelligence is an 8-person company—5 of them engineers—building a platform where founders run companies with AI agents. Their flagship product, Cofounder, deploys a full team of agents handling engineering, marketing, finance, sales, and support. To ship it, the company had to use the same agents on their own infrastructure, and they quickly learned the bottleneck was not the agents but the cloud platform beneath them. General Intelligence averages 10 PRs and 70+ commits per engineer per day, with over 4,000 preview branches and roughly 100 parallel app versions running during a workday.
"If you want to make a system where an agent drives all of your infrastructure, you should probably be using Vercel." — Andrew Pignanelli, CEO
The infrastructure requirement for agentic development
A coding agent runs dozens of processes in parallel. It queries logs as data, parses error output as input, and needs every dashboard action as a programmatic API. The design constraint is total control: anything a human can do on the cloud platform must be exposed through a CLI or API—killing deployments, editing DNS, managing billing, changing configs. Most cloud providers fall short, and that's why General Intelligence moved off Render to Vercel.
Render worked for the early product, but provisioning full-stack preview environments was painful from the start. When the team began building Cofounder's CTO agent, Python support limitations became a blocker. "There are vendors that let an agent do 5% of the work, and vendors that let an agent do 50%," explained Pignanelli. "We needed a platform that would let an agent do 100%."
No local development, only previews
The migration changed how General Intelligence builds software. Engineers no longer develop locally. Every change made by the CTO agent goes to a Git branch, spins up a preview environment, and is tested end-to-end by a browser agent hitting the live URL.
"We don't configure local development at all. Each branch has a preview environment, and within 30 seconds I'm using that version of the app. That just doesn't exist on other platforms." — Andrew Pignanelli, CEO
The scale is substantial: at any point during a workday, around 100 versions of the app are running concurrently, each with its own preview environment. Much of that load is internal work—coding agents building business agents in Cofounder. Fluid compute usage grew 6.5x month-over-month, mostly from this internal engineering, on a token budget of roughly $5,000 per engineer per month.
General Intelligence was one of the first teams to move a complex full-stack Python backend to Vercel. Unifying on one platform reduced the surface area for both team and agent to a single CLI, API, and observability layer. When something breaks, the agent and the humans see the whole picture in one place.
Multi-tenant platform with managed infrastructure
Cofounder is built as a multi-tenant app using Vercel for Platforms. When a founder provisions a company, they get a real GitHub repository, a managed Vercel deployment, and their own domain with automated SSL and DNS—all through programmatic provisioning. The engineering agent inside each customer company is the same CTO agent General Intelligence uses internally, running the identical workflow: branch, preview environment, and browser agent verification on the live URL.
Looking ahead
The company plans to keep operating itself on the same products it ships to customers. As they add more departments and agents, the focus stays on the customer experience rather than cloud configuration.
"We don't want to worry about infrastructure for our users. We want them to be able to say 'make my app go' and have it work, and Vercel is the answer." — Andrew Pignanelli, CEO
Ninety percent of the company's SRE work is automated through Vercel and their own agents, and every Cofounder customer gets a managed Vercel account at launch. General Intelligence is building Cofounder as the first full-stack agent company platform.
